S1-1.6 Wednesday, Jan. 4 Dimensionality reduction to understand sensory influences on turning in large scale behavior in Drosophila CENSI, Andrea*; STRAW, Andrew D. ; SAYAMAN, Ros; MURRAY, Richard M.; DICKINSON, Michael H.; California Institute of Technology; Research Institute of Molecular Pathology (IMP), Vienna, Austria; California Institute of Technology; California Institute of Technology; Department of Biology, University of Washington, Seattle, Washington email@example.com
A prerequisite for the mechanistic understanding of how nervous systems orchestrate behavior is a phenomenological description of those behaviors, in terms of the relevant features of the external stimuli and the animal's internal state. This is particularly challenging in experiments where the animal experiences very rich stimuli and generates complex behaviors, as opposed to experiments where the stimulus and measured responses are highly constrained. Here we describe a new approach that models turn-making decision process of the fruit fly Drosophila using machine learning techniques applied to a large dataset of free-flight 3D movement trajectories through a large environment. Our technique reduces the dimensionality of a complex visual input stream to a single variable that best predicts the decision to turn and successfully much of the structure in turn statistics in the tested environment. This analysis places bounds on the relative importance of various factors contributing to the fly behavior. In particular, the statistics of searching behaviors have alternately been attributed as either reactions to exogenous stimuli or as the direct behavioral manifestation of an intrinsic stochastic source with statistical properties matched to the search task. Our analysis suggests that flies will predominantly use such sensory cues to guide search rather than relying predominantly on an endogenous stochastic source as an important behavioral mechanism. We argue that a similar relationship is likely to be found for any searching organism endowed with sensory abilities.